AI Agents
Long-horizon task behavior, tool use, planning, recovery, and context evolution.
Founding-stage nonprofit AI research
Open Research
for More Responsible
AI
RedMind Research Foundation is a founding-stage nonprofit research organization focused on AI agents, model evaluation, human-AI collaboration, and responsible deployment.
Founding phase: donations are not currently being accepted.
Open benchmark note
agent step 07 / context preserved
Mission
Research That Serves the Public
We study advanced AI systems in ways that make their capabilities, limits, and social consequences easier for the public to understand.
Evaluation
Benchmarks and public reports that make model strengths and failures inspectable.
Responsible Deployment
Practical methods for safety, reliability, transparency, and accountability.
Research
Careful studies for agentic, evaluated, human-centered AI.
RedMind's early research focus connects long-horizon agent behavior, practical model evaluation, and workflows where humans retain judgment and accountability.
Agents
Agentic Systems & Long-Horizon Reasoning
Studying how AI agents preserve goals, context, and accountability across multi-step work.
Benchmarks
Model Evaluation & Benchmarking
Designing open evaluations that reveal practical model behavior, not just leaderboard scores.
Collaboration
Human-AI Collaboration
Exploring workflows where people and AI systems share judgment, feedback, and responsibility.
Recent Outputs
Research artifacts and accepted work
Early outputs include accepted papers on agentic context evolution, verification, and socially grounded generative agents.
Accepted & Presented
ACL 2026 (SURGeLLM)
Graph-Regularized Agentic Context Evolution
Dan Hsu, Jingsyue Lin, Tang-Hsin Lin, Luke Lu*
Accepted
COLM 2026 (Context Beyond the Window)
Scoped Verification for Reliable Long-Horizon Agentic Context Evolution under Distribution Shift
Dan Hsu, Luke Lu*
Accepted
COLM 2026 (Workshop on Agent Behavior)
Evaluating Generative Agents with Actions Grounded in Socially Distributed Task Environments using Incognita
Dan Hsu, Luke Lu*
Governance Commitments
Human-governed, agent-collaborative, open-by-default.
RedMind's founding operating framework is designed so AI agents can support rigorous research and documentation while humans retain legal, financial, board, filing, donation, and public-representation authority.
Human Authority
Board members and officers retain responsibility for mission, compliance, conflicts, spending, filings, public claims, and formal organizational decisions.
Bounded Agent Work
Agents may research, monitor, summarize, draft, maintain internal trackers, and recommend options. They do not bind or externally represent the Foundation.
Open Research
Research outputs, methods, benchmarks, tools, and decision rationale should be public where appropriate, with exceptions for privacy, safety, legal, and confidential matters.
This is a founding-phase operating framework and may evolve as formal board policies are adopted.
Open by Design
Public knowledge should stay public.
RedMind intends to publish research, tools, datasets, benchmarks, and educational resources openly for public benefit where appropriate. The work is intended to support students, researchers, builders, policymakers, and the broader public.
Open publications
Public datasets
Reproducible benchmarks
Educational resources
Founding Phase
Help Shape the Future of AI Research
Follow independent research intended to make advanced AI more useful, trustworthy, and broadly accessible.
Donations are not currently being accepted while IRS review, banking, and charity compliance steps are pending.
Contact
Start a research conversation.
Reach out about research collaboration, governance, educational resources, or future support once IRS review, banking, and donation readiness are complete.